Highlights
-
• Foreign language effect on moral judgments extends to informally acquired languages
-
• Utilitarian agents judged more moral and braver in foreign versus native language
-
• Bravery–morality relationship strengthens in foreign language contexts
-
• Testing in native versus foreign language can significantly distort cross-cultural findings
Moral psychology researchers often present participants with sacrificial dilemmas to understand moral decision-making. These dilemmas can be classified as direct or indirect (DiIanni, Reference DiIanni1977). In personal dilemmas, like the footbridge case, an agent must directly harm someone (e.g., push a person onto train tracks) to save multiple lives. In impersonal dilemmas, like the switch case, the harm is more indirect (e.g., pulling a lever to redirect a trolley). Personal dilemmas typically elicit stronger emotional reactions and are judged as less morally acceptable, reflecting the doctrine of double effect – active harm is prohibited, but harm as a side-effect of a positive action is more acceptable (Foot, Reference Foot1967; Quinn, Reference Quinn1989).
While most research asks participants what they would do in such situations, people have little experience with life-or-death decisions. In contrast, forming moral judgments about others is something people do routinely. These judgments form rapidly – within milliseconds – and prove resistant to change (Willis & Todorov, Reference Willis and Todorov2006). Moreover, moral character evaluations strongly influence impression formation (Goodwin et al., Reference Goodwin, Piazza and Rozin2014; Wojciszke et al., Reference Wojciszke, Bazinska and Jaworski1998). The distinction between judging others’ (moral) conduct versus predicting one’s own actions is crucial (Białek, Turpin, et al., Reference Białek, Turpin and Fugelsang2019; Bialek & Grossmann, Reference Bialek and Grossmann2022). Critics note that responses to hypothetical dilemmas either do not predict actual behaviors (Bostyn et al., Reference Bostyn, Sevenhant and Roets2018) or do so only weakly (Plunkett & Greene, Reference Plunkett and Greene2019). These considerations suggest that studying moral impressions of agents making sacrificial decisions may be more ecologically valid than asking about one’s own hypothetical actions.
Sacrificial dilemma responses depend on cost–benefit analysis and the salience of moral rules that must be transgressed, but other factors also matter. One example is the language in which people encounter moral dilemmas. This phenomenon is called the moral foreign language effect (mFLE). It describes how people make more utilitarian judgments in a foreign language than in their native language, particularly in scenarios involving direct harm (Circi et al., Reference Circi, Gatti, Russo and Vecchi2021; Stankovic et al., Reference Stankovic, Biedermann and Hamamura2022).
Sacrificial dilemmas activate two types of moral intuitions: deontological (“do not harm”) and utilitarian (“maximize good”). Using a foreign language decreases the accessibility of both intuitions (Barabadi et al., Reference Barabadi, Tabar and Booth2023; Białek, Paruzel-Czachura, et al., Reference Białek, Paruzel-Czachura and Gawronski2019; Hayakawa et al., Reference Hayakawa, Tannenbaum, Costa, Corey and Keysar2017; Hennig & Hütter, Reference Hennig and Hütter2021; Muda et al., Reference Muda, Niszczota, Białek and Conway2018). This essentially “un-moralizes” the problem (Muda et al., Reference Muda, Niszczota, Hamerski and Białek2025). However, mFLE often produces no main effect in large and diverse sets of moral dilemmas because changes in different intuitions cancel each other out across varied scenarios. The decrease in deontological inclination is more robust and salient in high-conflict moral dilemmas, the type most commonly used by researchers. Some authors view this as the main driver of the foreign language effect, linking foreign language use to increased utilitarian choice (Stankovic et al., Reference Stankovic, Biedermann and Hamamura2022).
This mechanism described above may seem complex, so we illustrate it using a hypothetical example where deontological and utilitarian inclinations can be quantified on a scale from 0 to 10 (Figure 1). In indirect dilemmas, deontological and utilitarian inclinations are typically activated with similar strength – both at approximately 8 points. This creates a balanced conflict with no difference between the two inclinations (Δ = 0), leading participants to endorse utilitarian and deontological solutions with roughly equal frequency. When participants evaluate the same dilemma in a foreign language, both inclinations are reduced proportionally. Assuming a 25% decrease in activation strength, both inclinations drop to 6 points while maintaining their balance (Δ = 0). Since the reduction affects both inclinations equally, decision patterns remain largely unchanged.
Moral foreign language effect in direct and indirect dilemmas.
Direct dilemmas require actively causing harm to another person. This morally repulsive nature triggers stronger deontological inclinations compared to indirect dilemmas. Since the consequences are identical in both dilemma types, utilitarian inclinations remain similarly activated. In our example, deontological inclinations reach 10 points while utilitarian inclinations remain at 8 points, creating a 2-point advantage for deontological considerations (Δ = 2). This deontological advantage explains why fewer people endorse harmful actions in direct dilemmas compared to indirect dilemmas. When evaluated in a foreign language, both inclinations are again reduced by 25%. However, this proportional reduction has differential effects: the stronger deontological inclination decreases from 10 to 7.5 points (a 2.5-point reduction), while the utilitarian inclination decreases from 8 to 6 points (a 2-point reduction). Although both inclinations weaken proportionally, the absolute reduction is larger for the initially stronger deontological inclination. This reduces the deontological advantage from 2 to 1.5 points (Δ = 1.5), resulting in more utilitarian judgments when direct moral dilemmas are evaluated in a foreign language.
This model explains why foreign language effects are more pronounced in direct dilemmas than indirect moral dilemmas. The effect does not stem from foreign language specifically enhancing utilitarian reasoning, but rather from its greater absolute impact on the stronger deontological inclinations that typically dominate direct moral dilemmas.
Earlier in its development, mFLE researchers proposed several explanations for this effect: decreased emotional reactions, increased cognitive engagement and increased psychological distance (Hayakawa et al., Reference Hayakawa, Costa, Foucart and Keysar2016). These explanations are complementary rather than competing with the one described above in a way that the latter mechanisms may explain the former using more basic-level processes.
The mFLE has been primarily documented in WEIRD (Western, Educated, Industrialized, Rich, Democratic; Henrich et al., Reference Henrich, Heine and Norenzayan2010) societies where foreign languages are typically learned through formal education. Moral judgments in non-WEIRD populations often differ from Western patterns. Tanzanians (Smith & Apicella, Reference Smith and Apicella2022) and Nicaraguans (Winking & Koster, Reference Winking and Koster2021) have shown more utilitarian tendencies, while Papuans demonstrated more deontological responses (Sorokowski et al., Reference Sorokowski, Marczak, Misiak and Białek2020). Recent studies suggest the mFLE can also occur with local dialects (Miozzo et al., Reference Miozzo, Navarrete, Ongis, Mello, Girotto and Peressotti2020; Peressotti et al., Reference Peressotti, Lorenzoni and Miozzo2023, Reference Peressotti, Pianezzola, Battistutta and Miozzo2024) or foreign languages learned alongside local languages (Stibbard-Hawkes et al., Reference Stibbard-Hawkes, Abarbanell, Mabulla, Endeko, Legare and Apicella2024).
In Tanzania, multilingualism is common, with distinct linguistic contexts that differ from WEIRD societies. In our study, the native language is Chagga (also known as Mochi), a stable local language belonging to the Niger–Congo language family (Eberhard et al., Reference Eberhard, Simons and Fenning2015). It serves as the first language for everyone within the ethnic community, but is not taught in schools. The foreign (second) language is Swahili, also part of the Niger–Congo language family and is Tanzania’s official national language. The informal way of acquiring both languages creates a unique context for studying the mFLE because other research typically contrasts an informally acquired native language with a formally acquired foreign language (e.g., taught at school).
Our study makes two key contributions. First, we examine the mFLE in this distinctive multilingual context where foreign language acquisition occurs informally rather than through classroom instruction. Second, we focus on moral impressions (judgments of morality and bravery) of others rather than one’s own hypothetical actions, potentially offering greater ecological validity given that people regularly form moral judgments about others but rarely face life-or-death moral dilemmas themselves.
We investigated how Chagga speakers evaluate the morality and bravery of utilitarian agents in sacrificial moral dilemmas presented in either Chagga (native) or Swahili (foreign). We hypothesized that utilitarian agents would be rated more positively when evaluated in Swahili compared to Chagga, especially when described as acting in a direct dilemma.
1. The Experiment
The study was not preregistered. All data and materials are available at: https://doi.org/10.17605/OSF.IO/JBTZ3
2. Method
2.1. Participants
The final sample comprised 103 native speakers of Chaga (40 female, 63 male), with a mean age of 35.9 years (range: 18–82). The sample size was determined by logistical and financial constraints rather than a priori power calculations. Sensitivity analysis indicated that this sample provided 80% power to detect between-subject effects of d = 0.56 or larger. Consequently, null findings should be interpreted with appropriate caution due to potential Type II errors for smaller effect sizes.
2.2. Procedure and materials
Data collection occurred through one-on-one interviews in local communities. Each participant evaluated two vignettes featuring culturally adapted moral dilemmas: one involving direct harm and one involving indirect harm to save five people. The dilemmas were presented in counterbalanced order to control for sequence effects. Materials were originally designed in English and translated into either Chagga or Swahili by a local interpreter. The interpreter was experienced in conducting similar cross-cultural research and blind to the hypothesis, but the accuracy of his translation was not independently verified.
Participants were randomly assigned to evaluate the scenarios either in their native or foreign language. Rather than asking about their own hypothetical actions, we presented participants with culturally adapted scenarios where a male agent had already decided to act utilitarianly (sacrificing one to save five). The two scenarios differed in the directness of the action required to save five other people from being killed by a stampede of wildebeests: indirectly by guiding the stampede away toward one villager, or more directly by asking an unaware person to make a loud noise to attract the stampede toward himself. Participants rated this agent on morality and bravery using 10-point Likert scales, with higher values indicating greater morality/bravery.
We also asked about participants’ willingness to gamble; however, due to coding issues, these results showed irregular patterns and were excluded from analysis. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All study materials and data are available at: https://doi.org/10.17605/OSF.IO/JBTZ3
3. Results
Using JAMOVI 2.6.44 (The Jamovi Project, 2021), we conducted two general linear mixed models: one with the dependent variable of morality and another with bravery. In each, the clustering variable was participant ID, and the fixed effect was language and dilemma type, coded as −0.5 for FL and indirect dilemma and 0.5 for NL and direct dilemma. The descriptive statistics are presented in Figure 2.
Bravery and morality as a function of language of presentation.
3.1. Morality ratings
The regression model was significant, F(3, 202) = 6.80, p < .001, predicting about 9% of the variance in morality ratings. Utilitarian agents were rated as significantly more moral when evaluated in the foreign language (Swahili) compared to the native language (Chagga), B = −1.15, 95% CI [−1.80, −0.50], p < .001, regardless of the dilemma type, B = −0.56, 95% CI [−1.85, 0.73], p = .394. Additionally, agents making indirect sacrifices were judged as more moral than those making direct sacrifices (B = −0.90, 95% CI [−1.54, −0.25], p = .007), consistent with previous research on the doctrine of double effect.
3.2. Bravery ratings
The regression model was significant, F(3, 202) = 3.71, p = .013, predicting about 5% of the variance in bravery ratings. Although the pattern descriptively resembled morality ratings, dilemma type did not significantly affect bravery judgments, B = −0.57, 95% CI [−1.26, 0.11], p = .098. Nevertheless, the foreign language effect was clearly present – utilitarian agents were rated as braver when their actions were evaluated in Swahili rather than Chagga, B = −0.99, 95% CI [−1.67, −0.31], p = .005, regardless of the dilemma type, B = −0.26, 95% CI [−1.63, 1.10], p = .703.
3.3. The interplay between bravery and morality
We explored whether the relationship between bravery and morality differed by language by regressing morality ratings on bravery ratings and language (coded as −0.5 and 0.5), separately for direct and indirect dilemmas. We speculated that deontological choice can be seen as an act of cowardice simply because it is passive, whereas utilitarian choice is an active intervention in reality. Thus, morality ratings could be increased simply because of the action/inaction asymmetry in utilitarian/ deontological choice. This relationship between bravery and morality, and more prominently its moderation by language, is purely exploratory analysis not driven by any established theories.
For direct dilemmas, we found significant effects of bravery (β = 0.42, p < .001), language (β = 0.40, p < .001) and their interaction (β = 0.70, p < .001). Decomposing this interaction revealed that bravery strongly predicted morality in the foreign language (β = 0.77, p < .001) but not in the native language (β = 0.07, p = .602).
For indirect dilemmas, bravery predicted morality (β = 0.46, p < .001), while language (β = 0.21, p = .208) and the language–bravery interaction (β = 0.31, p = .087) were not significant. Despite the non-significant interaction, we conducted exploratory decomposition given our limited statistical power. This revealed a similar pattern to direct dilemmas: bravery predicted morality more strongly in the foreign language (β = 0.61, p < .001) than in the native language (β = 0.30, p < .001).
4. Discussion
Our study with over 100 Tanzanians revealed that utilitarian moral agents were rated as more moral and braver when described in Swahili (foreign language) compared to Chagga (native language). These effects were substantial, approaching one point on a 10-point scale. Moreover, bravery was a stronger predictor of morality in the foreign language condition. These findings complement Stibbard-Hawkes et al.’s (Reference Stibbard-Hawkes, Abarbanell, Mabulla, Endeko, Legare and Apicella2024) study with the Hadza community, which found a descriptively similar but non-significant increase (6.2%, 95% CI [−0.45, 12.8]) in willingness to make sacrificial decisions when dilemmas were presented in a foreign language.
Our results demonstrate that moral impressions of others follow the same trajectory as one’s own predicted actions or judgments of permissibility – all become more affirming utilitarianism in a foreign language. This seemingly fits to a pattern reported in meta-analyses of the mFLE, showing a robust effect of increased utilitarian choices of g = 0.22 (Circi et al., Reference Circi, Gatti, Russo and Vecchi2021). Because people trust more agents who are like-minded, that is utilitarians prefer agents making utilitarian choices (Chandrashekar et al., Reference Chandrashekar, Bostyn, De Clercq and Roets2025), mFLe-driven utilitarian judgments could translate into a preference for utilitarian agents. Yet, the observed pattern could result from generally more lenient evaluations of moral transgressions in a foreign language (Woumans et al., Reference Woumans, Van der Cruyssen, Van Cruyssen and Duyck2020). In short, the required sacrifice of a single innocent person may be less wrong, thus improving its relative prominence to deontological considerations. However, an important caveat exists: we only examined perceptions of utilitarian agents. Hence, we cannot disentangle the increase in utilitarian inclination from merely more lenient moral evaluations.
The stronger relationship between bravery and morality in the foreign language condition suggests three potential mechanisms: (1) reduced emotional granularity in foreign language processing (Pavlenko, Reference Pavlenko2012) may limit access to nuanced moral intuitions, causing greater reliance on salient traits like bravery; (2) psychological distance created by a foreign language (Xu et al., Reference Xu, Wang, Zhu, Liao, Wang, Wang and Liu2024) may strengthen conceptual associations between bravery and morality as aspects of a unified “virtuous character” or (3) Swahili might activate different cultural scripts about the relationship between courage and morality. Notably, the third explanation is the least promising simply because its local character, which cannot be extrapolated to other findings in the culturally invariant findings on the foreign language effect.
Our findings have methodological implications for cross-cultural moral psychology. As researchers like Awad et al. (Reference Awad, Dsouza, Kim, Schulz, Henrich, Shariff, Bonnefon and Rahwan2018) or Graham et al., (Reference Graham, Haidt, Koleva, Motyl, Iyer, Wojcik, Ditto, Devine and Plant2013) attempt to develop universally acceptable moral codes, the language of assessment emerges as a critical methodological factor that can significantly influence results and potentially lead to misunderstandings. Researchers conducting cross-cultural studies should carefully consider how the language of assessment might influence participants’ moral judgments, especially in multilingual populations.
Key limitations include the lack of assessment of participants’ language proficiency and examining only male utilitarian agents. Future research should: (1) include both utilitarian and deontological agents to determine the specificity of the language effect; (2) measure language proficiency and age of acquisition to examine whether the effect varies with fluency and (3) explore similar effects across different moral violation types and (4) potential interactions between the sex of participants and the potential victims.
While our findings are consistent with the moral foreign language effect observed in WEIRD populations, the specific linguistic context in Tanzania – where Swahili is acquired informally rather than through formal education – suggests that the effect extends to any non-native language regardless of acquisition method. This broadens our understanding of how language shapes moral cognition across diverse cultural and linguistic contexts.
Data availability statement
The data that support the findings of this study are openly available in the Open Science Framework at https://doi.org/10.17605/OSF.IO/JBTZ3
Author contribution
Conceptualization: M.Bi.; funding acquisition: M.Bi. and P.S.; supervision: M.Bi.; writing-review and editing: A.B., M.Bu. and P.S.; data curation: A.B. and W.M.; project administration: M.Bu., P.S. and W.M.; visualization: W.M.; writing-original draft: M.Bi., H.K.T. and W.M.; investigation: M.Bu.; methodology: M.Bi., M.Bu. and P.S.
Funding statement
The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This research was supported by grant no. 2020/38/E/HS6/00282 from the National Science Centre (Poland) to Michał Białek. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Competing interests
The authors declare none.
Open access
